Energy Efficiency in Cloud Computing · techniques and algorithms proposed to achieve efficient energy utilization in cloud computing, these techniques include: VM Migration, Consolidation

Abstract—Cloud computing is one of the recent emerging

technologies that provides services to consumers in a pay as you

go model. Cloud computing offers ITC based services over the

internet and the use of virtualization allows it to provide

computing resources. Data Centers are the core of cloud

computing, which consists of: networked servers, cables, power

sources, etc. which host the running applications and store

Business information. High performance has always been the

most critical concern in cloud data centers, which comes at the

cost of energy consumption. The vital challenge is balancing

between system performance and power consumption by

reducing energy consumption without prejudicial impact on the

performance and quality of services delivered. There are many

techniques and algorithms proposed to achieve efficient energy

utilization in cloud computing, these techniques include: VM

Migration, Consolidation and Resources orchestration in cloud

computing. This paper provides a survey of approaches and

techniques for energy efficiency in cloud computing.

Index Terms—Cloud computing, energy efficiency, resource

management, virtualization.

I. INTRODUCTION

The progress of technology and incorporating networks,

storage and processing power led to new era of computing,

called cloud computing or commonly known as the cloud.

Cloud computing is defined as a technological paradigm that

allows on-demand access via the internet to a common shared

computing resources. It is considered to be a model for

supervision, storing and processing data online via the

internet [1]. Some cloud computing characteristics include

on-demand services, network access by using internet as a

medium, shared resources by pooling resources together to be

used by multiple clients and scalability by maintaining

elasticity of resources. Cloud computing offers different

services based on three delivery models, namely:

1) Software as a service (SaaS): this allows users of cloud to

access the providers apps (PA) over the internet.

2) Platform as a Service (PaaS): this allows users to deploy

their apps on a platform which service provider of cloud

(SPC) provides. 3) Infrastructure as a Service (IaaS): this allows users to rent,

store, process in an infrastructure provided by SPC.

The rapid growth in mobile devices and the storage needs

due to the adoption of cloud data networking are creating

huge data traffic due to the emerging issues of data centers

and also digital content, media and technology. Energy

consumption by the organizations that provide cloud service

is continuously increasing. It has been concluded that the

Manuscript received July 14, 2018; revised October 14, 2018.

The authors are with the British University, Egypt (e-mail:

[email protected], [email protected],

[email protected], [email protected],

[email protected]).

amount of energy consumed by the data centers of the cloud

service providers is equal to 1.5% of power supplied to an

entire city [2]. The data centers for cloud service providers

are used for hosting the cloud applications which are

normally consuming massive amounts of resources that

utilize a huge percentage of electrical energy, which produces

growth in operational cost and results in emission of Co2 [3].

Cloud service providers ensure reliability and load balancing

for the services provided to the users around the world by

keeping servers operating all the time. In order to satisfy the

Service Level Agreement (SLA), cloud service providers has

to supply power continuously to data centers, which utilizes a

huge amount of energy by the data center and subsequently

increases the cost of investment. Thus, it has been noticed

that high performance has been the sole concern in data

center deployments. This demand has been achieved without

paying attention to the amount of energy consumed. The key

challenge is to balance between system performance and the

power consumption. It was found that a huge amount of

energy is consumed due to idle and overloaded servers in data

center. According to [4], idle servers use 69-97% of total

energy in the presence of enabled power management

function. This paper will present an overview of the different

methodologies to have energy efficiency in cloud by

introducing some of the current proposed solutions as servers

load balancing, VM virtualization, VM migration and

resource allocation.

A. Clouds

In order to meet the rapid-changing business and

organization needs, organizations need to devote budget and

time to accelerate up their IT infrastructure such as software,

hardware and network services. Regardless of the utilization

of on-site IT framework, scaling the system could be difficult.

And also the organizations are often incapable of achieving

an ideal use of IT foundation. Thus, the cloud computing is

the proposed solution. According to National Institute of

Standards and Technology (NIST), cloud computing is the

delivery of IT resources on-demand utilization by providing a

pay as you go model for the consumers, while you can

self-serve for the services that you need to your own

application or any IT infrastructure that you need. A cloud

computing service consists of highly utilized resources

including software applications or virtual storages that can be

used upon user request, consumers can simply connect to the

cloud and use the available resources. This causes

organizations to stay away from capital consumption for

on-premises framework assets and scaling up or downsizing

according to business requirements [3]. Cloud computing

services can be deployed using three different models a

private cloud, public cloud or a hybrid cloud. Private cloud

function solely for one organization on a private network and

is its highly secure. Public cloud is owned by the cloud

service provider and offers the highest level of efficiency and

Energy Efficiency in Cloud Computing

Mohamed Deiab, Deena El-Menshawy, Salma El-Abd, Ahmad Mostafa, and M. Samir Abou El-Seoud

International Journal of Machine Learning and Computing, Vol. 9, No. 1, February 2019

98doi: 10.18178/ijmlc.2019.9.1.771

shared resources and hybrid cloud is considered to be a

combination of private and public deployment models. In a

hybrid cloud, specific resources are run or used in a public

cloud and others are run or used on-premises in a private

cloud this provides increased efficiency. Fig. 1 illustrates the

architecture of cloud computing [3].

Fig. 1. A high level system architecture of cloud computing [3].

B. Data Centers

Data centers provide an IT backbone for cloud computing.

A data center is a technical facility that houses organizations

IT operations and equipment where it stores, manages and

disseminates its data. A data center houses and networks

most critical systems and are vital to business continuity and

operations. A Data center is considered to be the heart of

cloud computing which contains all the cloud resources

including servers, network, cables, etc. on which business

information is stored and applications run. Until recently,

high performance has been the sole concern in data center

distributions and this demand has been satisfied without

paying much attention to energy consumption. However, an

average data center consumes as much as the consumption of

25,000 houses [3]. As the energy availability decreases and

energy cost proportionally increases, the need for shifting the

focus for utilizing data center resource management to

optimize energy performance while maintaining service

performance is becoming a necessity. Thus cloud service

providers need to adjust their energy measures to ensure that

their profit margin is not dramatically reduced due to high

energy costs.

II. ENERGY EFFICIENT COMPUTING

Energy saving techniques in computing equipment have

been classified as static power management (SPM) and

Dynamic power management (DPM). SPM and DPM are

completely different in categorization, SPM are more energy

efficient at single system and supposed to be under the

category of hardware level techniques, and since SPM

techniques are related to hardware level efficiency, low

power consumption circuit designing is an example of this

technique. On the other side, DPM are more energy efficient

in large systems and supposed to be under the category of

level resource management methods. Also, DPM techniques

are mostly implemented in software or on network layer, for

example protocol design and algorithms. Fig. 2 shows an

overview of various energy management schemes in

computing equipment. Energy aware scheduling, energy

efficient routing, load balancing, virtualization, resource

consolidation and migration. Since high availability as well

as quality of service and performance guarantee are still

ignored which is most required in such distributed

environments as the customers pay for their provisioned

resources. The customers would not pay or may switch to

other similar service providers if either quality of service or

expected performance level is not achievable. Energy issues

are supposed to be critical and also needs to be managed

properly in some environment where mobile cloud

computing is involved. Reducing the amount of energy used

by applications through green compilers and robust

programming can be achieved through application/software

level methods [5]. In next sections, Application level and

high level resource management techniques are discussed to

achieve energy efficiency in single system, clusters, grids and

cloud datacenters.

Fig. 2. Overview of various energy management schemes in computing

equipment [6].

III. RESOURCE SCHEDULING MODEL FOR ENERGY SAVING

IN CLOUD COMPUTING

Basically the resource model of the cloud data center and

the dynamic power model of the physical machine are both

built, and afterward a three-dimensional virtual resource

scheduling method (TVRSM) is proposed along with related

algorithms [7]. The process of virtual resource scheduling in

TVRSM is divided into three stages, these stages are virtual

resource allocation stage, virtual resource scheduling stage

and virtual resource optimization stage. Regarding TVRSM,

three algorithms are designed corresponding to the

mentioned stages of the virtual resource scheduling. These

algorithms are MVBPP based heuristic virtual resource

allocation algorithm (HVRAA), multi-dimensional power

aware based virtual resource scheduling algorithm


99

(MP-VRSA) and virtual resource optimization algorithm

(VROA).

Initially, the first stage in TVRSM which is virtual

resource allocation stage is basically in charge of allocating

the requested VMs by the customer to the suitable hosts. This

stage is treated as multi-dimensional vector bin packing

problem (MVBPP) and the MVBPP based heuristics virtual

resource allocation algorithm (HVRAA) is proposed to solve

it. In addition, the second stage which is virtual resource

scheduling stage is responsible for migrating the VMs from

the overload hosts to other hosts with lower resource

utilization by using the VM migration technology in order to

achieve load balancing of the cloud data centers and also to

minimize the amount of violations of Service Level

Agreement. The multidimensional power-aware based virtual

resource scheduling algorithm (MP-VRSA) is proposed in

this stage. Furthermore, the third stage which is virtual

resource optimization stage is in charge of migrating the VMs

from the hosts with the least resource utilization to other

hosts and switch the original hosts to sleep mode, this process

can further reduce the energy consumption of the cloud data

centers by designing the virtual resource optimization

algorithm (VROA). Finally, the authors verified the

effectiveness of the proposed method through

experimentation. The results prove that the TVRSM is able to

efficiently allocate and manage the virtual resources in the

cloud data center. And a comparison is made between the

proposed methods with other traditional algorithms. The

results showed that the TVRSM can effectively reduce the

energy consumption of the cloud data center and minimize

the amount of violations of Service Level Agreement.

A. Resources Model of Cloud Data Center

In [8], the authors proposed a resource model of the cloud

data center, which is shown in Fig. 3, it consisted of M

clusters, and each cluster contains N physical machines.

Several virtual machines are deployed on each physical

machine. According to the resources owned by the virtual

machine, each virtual machine can run multiple applications.

So the load of each virtual machine results from the

applications running on the virtual machine. The node

controller runs on each physical machine is responsible for

monitoring the resource utilization of each physical machine

and control the physical machines status such as setting the

physical machine to sleep mode or activating the sleeping

physical machine. Also, the node controller sends the

management commands to the Hypervisor in order to adjust

the resources owned by the local virtual machines. The global

resources management node is responsible for scheduling

and allocating all the resources owned by the cloud data

center. It can manage and monitor all the resources and

implement the load balance of the cloud data center. As

shown in Fig. 3, each physical machine is characterized by

the CPU performance, amount of RAM and network

bandwidth, and each physical machine can run multiple

virtual machines, and the physical resource owned by each

virtual machine consists of CPU, memory capacity and

network bandwidth. The physical machines use Network

Attached Storage (NAS) instead of having local disks. It uses

NAS in order to save data, which can ease the data sharing

between all physical machines and enable live migration of

virtual machines quickly.

Fig. 3. Resource model of CDC [7].

B. Dynamic Power Model

In [7], the authors have shown that the power consumption

by PMs can be described by a linear relationship between the

power consumption and CPU utilization. Hence, the power

consumption Pi (t) of PM i running on time t can be expressed

by the relationship between the CPU utilization ui (t) of PM i

on time t and the maximum power consumption PiMax of

PM i, as shown in the below formula:

𝑝𝑖(𝑡) = 𝑘 . 𝑃𝑖𝑀𝑎𝑥 + (1 − 𝑘) . 𝑃𝑖𝑀𝑎𝑥 . 𝑢𝑖(𝑡) (1)

As shown in formula 1, PiMax is considered to be the

maximum power consumption of host, Pi(t) is considered to

be the power consumption of PM i running on time t, k

considered to be the fraction of power consumption when the

host is in idle state and ui(t) is the CPU resource utilization of

the PM on time t. In the below formula, the CPU utilization of

the PM i is defined as the ratio of the total CPU resources

requested by the all VMs running on PMi on time t to the all

CPU resources owned by the PM I as shown in formula 2.

𝑢𝑖(𝑡) = ∑𝑐𝑝𝑢𝑟𝑗(𝑡)

𝐶𝑃𝑈𝑖𝑇𝑜𝑡𝑎𝑙

𝑉𝑀𝑆𝑖𝑗=1 (2)

IV. RESOURCE SCHEDULING MODEL FOR ENERGY SAVING

IN CLOUD COMPUTING

In HVRAA, the main objective of the algorithm is to

assign all the VMs requested by customers to the minimum

number of PMs. The core idea of HVRAA is as follows:

select the VM which has the largest Weight Dot Product

(WDP), Select all the VMs that can fit the host and finally, if

there is no any fit the current host, then start a new host until

all the VMs are assigned into the hosts.

In MP-VRSA, the main objective of this algorithm is to

further reduce the energy consumption by identifying and

detecting the overloading hosts. The MP-VRSA is composed

of four steps: detecting the overloading hosts, choose the

VMs that need to be migrated from the overloading hosts,

selecting new hosts for the VMs to be migrated and

implementing the migration operation for all the overloading


100

hosts. The first step in MP-VRSA is detecting the

overloading hosts where the overloading detection strategy is

used in order to find the overloading hosts in the CDC to

determine whether the VMs running on the host need to be

migrated. The below steps are essential for detecting the

overloading hosts; setting the utilization threshold and if the

CPU utilization of a host exceeds the threshold, then the

overloading host can be detected and some VMs running on

the host need to be migrated. The second step in MP-VRSA is

VM selection strategy, once the host has been detected

overload. Maximum Correlation (MC) VM selection strategy

is used for selecting VMs to migrate from the overloaded host.

The idea of Maximum Correlation (MC) VM selection

strategy is that the higher the correlation between the loads of

VMs running on a host, the higher the probability of the host

overloading. The CPU utilization of VM is considered as the

load of VM. So according to this idea, VMs to be migrated

that have the highest correlation of the CPU utilization with

other VMs are selected. The third step in MP-VRSA is VM

placement strategy where the main task of this strategy is to

select the suitable host for the VMs that are migrated from the

overloading hosts. However, when the VMs are reallocated

to other hosts it is bound to make the CPU utilization of the

hosts increased. So, the Minimum Power Increasing Strategy

(MPIS) is designed in order to place the VMs into hosts

quickly and reduce the energy consumption in the CDC. In

Virtual Resource Optimization Algorithm (VROA), this

algorithm migrates the VMs from the hosts with the least

resource utilization to other PMs, and switch the original host

to sleep mode. Therefore, it can reduce the energy

consumption of data centers. VROA consists of three main

steps; after the virtual resource scheduling step is finished,

the VROA will select the host PM lowest with the lowest

CPU utilization and attempts to migrate the VMs to other

hosts. Then, the system will set host PM lowest to sleep mode

when the VMs migrate to other hosts successfully. Finally, If

any of the VMs on host PM lowest cannot be migrated, then

the host is kept active and all the migration of VMs are

canceled.

V. ENERGY-AWARE VIRTUAL MACHINE MIGRATION FOR

CLOUD COMPUTING

The proposed technique proposes another methodology for

maintaining energy efficiency in cloud computing, by

migrating the maximally loaded virtual machines to the least

loaded active machine, while maintaining system

performance by performing a live migration of the virtual

machines to ensure that all the running applications will not

get disconnected during migration. The proposed technique

introduces a new methodology for improving resource

utilization levels based upon the bio-inspired Firefly

optimization technique to achieve energy efficiency in cloud

data centers. The achievability of the proposed technique has

been shown by executing the results by using the CloudSim

simulator.

A. Firefly Optimization (FFO) Algorithm

The Firefly Optimization (FFO) algorithm has been

introduced by Xin-She Yang in the late 2007 and 2008 at

Cambridge University [9]–[11]. It was implemented upon the

fireflies flashing characteristics and behavior, the

characteristics have been introduced as follows:

1) One firefly is attracted to the other fireflies regardless of

their sex as all fireflies are unisex

2) The attractiveness is proportionate to the brightness, thus

they both decrease as their distance increases and for any

two flashing fireflies, the less bright one will be attracted

near the brighter one

3) The brightness of a firefly is calculated using the

objective function to be optimized

The (FFO-EVMM) Technique introduces the idea of

migrating the most loaded VM from an active node which

satisfies minimum criteria for energy consumption, to

another active node that consumes the least energy. The

technique is implemented in four main parts as shown in Fig.

4:

1) Selection of source node

2) Selection of VMs

3) Selection of destination node

4) Distance updated values.

Fig. 4. Flow chart for FFO algorithm [12].

VI. RESULTS AND ANALYSIS

Fig. 5. VMs vs hosts [12].

The statistical results for the proposed FFO-EVMM

algorithm were compared with the ACO-based and


101

FFD-based algorithms, using the CloudSim toolkit simulator

which has captured that that FFO-EVMM technique runs less

number of active hosts and performs less number of virtual

machines migration in comparison to ACO and FFD-based

algorithms.

As it's shown with the less number of running hosts and

live migrations, FFO-EVMM requires lesser energy demand

comparing to FFD and ACO algorithms as it has been noticed

from Fig. 5 and 6.

Fig. 6. VMs vs energy consumption [12].

VII. CONCLUSION

Cloud computing is considered one of the most crucial

technologies that provide services to consumers in a pay as

you go model. It offers ITC based services over the internet

and the utilization of virtualization allows it to provide

computing resources. Data centers are the core of cloud

computing that store business information and host the

running applications. High performance has always been the

sole concern of all in data centers. This concern has been

managed without considering energy consumption and

performance. The challenge is to balance between power

consumption and system performance. Many techniques and

algorithms have been proposed to achieve adequate energy

utilization in cloud data centers. This paper provided a survey

of recent approaches and techniques for energy efficiency in

cloud computing.

REFERENCES

[1] R. Buyya, C. S. Yeo, and S. Venugopal, “Market-oriented cloud

com-puting: Vision, hype, and reality for delivering it services as

computing utilities,” in Proc. 10th IEEE International Conference on

High Performance Computing and Communications, 2008, pp. 5–13.

[2] A. Beloglazov and R. Buyya, “Optimal online deterministic algorithms

and adaptive heuristics for energy and performance efficient dynamic

consolidation of virtual machines in cloud data centers,” Concurrency

and Computation: Practice and Experience, vol. 24, no. 13, pp. 1397–

1420, 2012

[3] A. Beloglazov, R. Buyya, Y. C. Lee, and A. Zomaya, “A taxonomy and

survey of energy-efficient data centers and cloud computing systems,”

Advances in Computers, Elsevier, 2011, vol. 82, pp. 47–111.

[4] N. R. D. Council. (2014). Scaling up energy efficiency across the data

center industry. [Online]. Available: https://www.infrastructureusa.org

[5] F. Fakhar, B. Javed, R. Rasool, O. Malik, and K. Zulfiqar, “Software

level green computing for large scale systems,” Journal of Cloud

Computing: Advances, Systems and Applications, vol. 1, no. 1, p. 4,

2012.

[6] M. Zakarya and L. Gillam, “Energy efficient computing, clusters, grids

and clouds: A taxonomy and survey,” Sustainable Computing:

Informatics and Systems, vol. 14, pp. 13–33, 2017.

[7] W. Zhu, Y. Zhuang, and L. Zhang, “A three-dimensional virtual

resource scheduling method for energy saving in cloud computing,”

Future Generation Computer Systems, vol. 69, pp. 66–74, 2017.

[8] R. Jhawar, V. Piuri, and P. Samarati, “Supporting security

requirements for resource management in cloud computing,” in Proc.

2012 IEEE 15th International Conference on Computational Science

and Engineering (CSE), IEEE, 2012, pp. 170–177.

[9] X.-S. Yang, Nature-Inspired Metaheuristic Algorithms, Luniver press,

2010.

[10] Firefly algorithms for multimodal optimization, presented at

International Symposium on Stochastic Algorithms, Springer, 2009, pp.

169–178.

[11] X.-S. Yang and X. He, “Firefly algorithm: Recent advances and

appli-cations,” International Journal of Swarm Intelligence, vol. 1, no.

1, pp. 36–50, 2013.

[12] N. J. Kansal and I. Chana, “Energy-aware virtual machine migration

for cloud computing-a firefly optimization approach,” Journal of Grid

Computing, vol. 14, no. 2, pp. 327–345, 2016.

M. Samir Abou El-Seoud received his BSc degree

in physics, electronics and mathematics from Cairo

University in 1967, his higher diplom in computing

from Technical University of Darmstadt (TUD) /

Germany in 1975 and his doctor of science from the

same University (TUD) in 1979. Currently, his

research interests are focused on computer aided

learning, parallel algorithms, mobile applications,

augmented reality, cloud computing, IoT, numerical

scientific computations and computational fluid mechanics. Professor

El-Seoud helds different academic positions at TUD Germany. Letest

full-professor in 1987. outside Germany professor El-Seoud spent different

years as a full-professor of computer science at SQU – Oman, Qatar

University, and PSUT-Jordan and acted as a head of computer science for

many years. At industrial institutions, Professor El-Seoud worked as

Scientific Advisor and Consultant for the GTZ in Germany and was

responsible for establishing a postgraduate program leading to M.Sc. degree

in Computations at Colombo University / Sri-Lanka (2001 – 2003). He also

worked as Application Consultant at Automatic Data Processing Inc.,

Division Network Services in Frankfurt/Germany (1979 – 1980). Professor

El-Seoud joined The British University in Egypt (BUE) in 2012. Currently,

he is Basic Science Coordinator at the Faculty of Informatics and Computer

Science (ICS) at BUE. Professor El-Seoud has more than 150 publications in

international proceedings and international reputable journals.

Ahmad Mostafa is an assistant professor at The

British University in Egypt. He graduated with his

Ph.D. from the the Center for distributed and mobile

computing at the University of Cincinnati, Ohio. His

research interests include routing, energy

conservation and localization in wireless sensor

networks, VoIP and physical implementation over

wireless mesh network, QoS in vehicular networks

as well as heterogeneous networks. His research

experience is reinforced by strong passion for education as he has taught

various courses at the British University in Egypt, the University of

Cincinnati and other universities in the US over the past eight years.


102

Documents

Energy Efficiency in Cloud Computing · techniques and algorithms proposed to achieve efficient energy utilization in cloud computing, these techniques include: VM Migration, Consolidation